Model-Based Analysis in a Relational Database

1. A computer-implemented method for model-based analysis in a relational database, the method comprising: accessing a model repository comprising a plurality of models respectively being adapted to perform, when used by an analytical program, a computational task, the model repository comprising a best-model, the best-model being a default model for performing the computational task, and the analytical program being a plug-in or add-on or an integral component of a database management system (DBMS) that maintains the relational database; creating a first database table in the relational database, the first database table having a predefined table structure that corresponds to the analytical program; reading the best-model from the model repository and storing the best-model in the first database table; receiving, by the analytical program, a request of a client device to perform the computational task, the request comprising input data; the analytical program determining if the request comprises a model-ID; if the received request does not comprise a model-ID, the analytical program reading the best-model currently stored in the first database table and using the read best-model for performing the computational task on the input data; if the received request comprises a model-ID, the analytical program: creating a second database table in the database, the second database table having the predefined table structure; reading a model associated with the model-ID from the model repository; storing the read model associated with the model-ID in the second database; and using the model read from the second table for performing the computational task on the input data; and the analytical program returning a result of the performing of the computational task to the client device, wherein the computational task is the classification of bioactive compounds, wherein the plurality of models are support-vector-machine-models and wherein the best-model is the one of the plurality of models which provides a most accurate classification.

2. The computer-implemented method of claim 1 , wherein the models in the model repository comprise different numbers and/or types of hyperparameters, the data format of the models in the model repository is the same for all models irrespective of the number and type of hyperparameters contained in a model, and the predefined table structures of the first and second database tables depend upon the number and/or type of hyperparameters in the models.

3. The computer-implemented method of claim 1 , wherein the computational task is the classification of bioactive compounds, wherein the plurality of models are support-vector-machine-models and wherein the best-model is the one of the plurality of models which provides a most accurate classification.

4. The computer-implemented method of claim 1 , the first database table being created and a copy of the best-model being stored in the first database table before, during or after the instantiation of an analytical program and before the request is received.

5. The computer-implemented method of claim 1 , each of the models in the model repository associated with a respective table structure selectively adapted to store said model in a structured format that corresponds to structural requirements of the analytical program, the first database table being created in accordance with the table structure assigned to the best-model, the creation of the second database table comprising: in response to receiving the request, dynamically identifying a table structure assigned to the one of the models identified by the received model-ID, and creating the second database table in accordance with the identified table structure.

6. The computer-implemented method of claim 1 , each of the models in the model repository associated with a model type, each of the model types associated with a table structure selectively adapted to store models of said model type in a structured format that corresponds to the analytical program, the first database table being created in accordance with the table structure associated with the model type to which the best-model is associated, the creation of the second database table comprising: in response to receiving the request, dynamically identifying the model type of the model indicated in the request, identifying the table structure associated with the identified model type, and creating the second database table in accordance with the identified table structure.

7. The computer-implemented method of claim 1 , further comprising: automatically creating the plurality of models by applying a machine learning logic on training data; automatically analyzing the plurality of models for identifying the best-model, the best-model being the one of the plurality of models being the best suited for solving the computational task; and storing all created models in the model repository.

8. The computer-implemented method of claim 1 , the plurality of models comprising multiple models having a different number of model parameters, wherein the table structure used for creating the second database table in response to a request comprising a model-ID of one of said multiple models depends on the number of model parameters and differs for each of the multiple models.

9. The computer-implemented method of claim 1 , the relational database being maintained by a DBMS, the analytical program comprising analytical SQL routines or a call to analytical SQL routines, the performing of the computational task comprising executing, by the DBMS, the analytical program routines, the analytical SQL routines being adapted to access specific columns of the first or second database table from which the model used by the analytical program is read.

10. The computer-implemented method of claim 1 , the relational database being maintained by a DBMS, the analytical program comprising analytical SQL routines or a call to analytical SQL routines, the method comprising: storing the received input data in one or more database tables of the relational database; wherein the performing of the computational task comprises executing, by the DBMS, the analytical program routines, the analytical SQL routines being adapted to access specific columns of the one or more database tables comprising the input data.

11. The computer-implemented method of claim 1 , the first database table being a static database table, the second database table being a temporary database table.

12. The computer-implemented method of claim 1 , the first database table being accessible by multiple client devices concurrently, the second database table being created within a session between a server computer and the client device and being inaccessible and invisible by other client devices.

13. The computer-implemented method of claim 1 , the relational database being maintained by a DBMS configured to repeatedly persist static tables in a non-volatile storage medium and to store temporary tables solely in a volatile storage medium.

14. The computer-implemented method of claim 1 , further comprising: receiving, by the analytical program, a further request of the client device to perform the computational task, the request comprising the input data and a model-ID of a further one of the models; the analytical program identifying the further model and a further predefined table structure selectively adapted to store the further model in a further structured format, creating a third table in accordance with the further predefined table structure, reading the identified further model from the model repository, storing the read further model in the third table and using the model read from the third table for performing the computational task on the input data; the analytical program returning a further result of the performing of the computational task with the further model to the client device.

15. A non-transitory hardware medium storing computer-executable program code, the program code executable by a computing system to: access a model repository comprising a plurality of models respectively being adapted to perform, when used by an analytical program, a computational task, the model repository comprising a best-model, the best-model being a default model for performing the computational task, and the analytical program being a plug-in or add-on or an integral component of a DBMS that maintains the relational database; create a first database table in a database, the first database table having a predefined table structure that corresponds to the analytical program; read the best-model from the model repository and store the best-model in the first database table; receive, by the analytical program, a request of a client device to perform the computational task, the request comprising input data; the analytical program to determine if the request comprises a model-ID; if the received request does not comprise a model-ID, the analytical program to read the best-model currently stored in the first database table and use the read best-model for performing the computational task on the input data; if the received request comprises a model-ID, the analytical program to: create a second database table in the database, the second database table having the predefined table structure, read a model associated with the model-ID from the model repository, store the read model associated with the model-ID in the second database table, and use the model read from the second database table for performing the computational task on the input data; and the analytical program to return a result of the performing of the computational task to the client device.

16. The medium of claim 15 , each of the models in the model repository associated with a respective table structure selectively adapted to store said model in a structured format that corresponds to structural requirements of the analytical program, the first database table being created in accordance with the table structure assigned to the best-model, the creation of the second database table comprising: in response to receiving the request, dynamically identifying a table structure assigned to the one of the models identified by the received model-ID, and creating the second database table in accordance with the identified table structure.

17. The medium of claim 15 , the program code further executable by a computing system to: receive, by the analytical program, a further request of the client device to perform the computational task, the request comprising the input data and a model-ID of a further one of the models; the analytical program identifying the further model and a further predefined table structure selectively adapted to store the further model in a further structured format, creating a third table in accordance with the further predefined table structure, reading the identified further model from the model repository, storing the read further model in the third table and using the model read from the third table for performing the computational task on the input data; the analytical program returning a further result of the performing of the computational task with the further model to the client device.

18. A computer system comprising a server computer system, the server computer system comprising: a hardware processor; a memory; a network interface adapted for connecting the server computer system via a network to at least one client device; a model repository, the model repository comprising a plurality of models respectively being adapted to perform, when used by an analytical program, a computational task, the model repository comprising a best-model, and the analytical program being a plug-in or add-on or an integral component of a DBMS that maintains the relational database; and a DBMS that maintains a relational database; the analytical program configured for: creating a first database table in the relational database, the first database table having a predefined table structure that corresponds to structural requirements of the analytical program; reading the best-model from the model repository and storing the best-model in the first database table receiving a request of the at least one client device to perform the computational task via the network, the request comprising input data; determining if the request comprises a model-ID; if the received request does not comprise a model-ID, reading the model currently stored in the first database table and using the read best-model for performing the computational task on the input data; if the received request comprises a model-ID, creating a second database table in the relational database, the second database table having a predefined table structure that corresponds to the analytical program, reading the identified model from the model repository, storing the read identified model in the second database table, and using the model read from the second database table for performing the computational task on the input data; and returning a result of the performing of the computational task to the client device via the network.

19. The computer system of claim 18 , further comprising the at least one client device and at least one further client device, wherein the analytical logic is configured to read the best-model from the first database table in response to respective requests from multiple ones of the client devices, and wherein analytical logic is configured to store and read models identified via model-IDs in a request in temporary tables that are created specifically for the session context of requests of individual ones of the client devices.